Wigner Semicircle Law in Random Matrices

• Wigner Semicircle Law is a universal spectral law that defines the eigenvalue density of large random matrices, typically supported on [-2, 2].
• It employs both the moment method and the resolvent approach to accurately capture eigenvalue statistics, ensuring robustness even with dependent or non-identically distributed entries.
• Extensions of the law demonstrate its relevance in analyzing random regular graphs, weighted networks, and random polynomials, impacting fields like combinatorics and mathematical physics.

The Wigner semicircle law describes the universal limiting spectral distribution for a broad class of large random matrices, particularly Wigner matrices, and has deep connections to combinatorics, probability theory, and mathematical physics. It characterizes the eigenvalue density in the limit of large matrix size, supported on a finite interval and given by the semicircular density function. The semicircle law exhibits remarkable robustness across models with independent, dependent, and even non-identically distributed entries, as well as in various generalized contexts such as regular graphs, Laplacians of networks, random polynomials, and higher-order tensors.

1. The Semicircle Law: Definition and Universal Regimes

For an $n \times n$ real symmetric (or Hermitian) Wigner matrix $W_n$ with independent, centered (mean-zero) entries and appropriately normalized variance (typically $\mathrm{Var}(W_{ij}) = 1/n$ for $i \neq j$), the empirical spectral distribution (ESD) of its eigenvalues converges, as $n \to \infty$, to the semicircle distribution

$$\rho_{sc}(x) = \frac{1}{2\pi}\sqrt{4 - x^2}\,\mathbf{1}_{|x| \leq 2}$$

supported on $[-2, 2]$. This convergence holds in probability and, under mild conditions on higher moments or weak correlation between entries, almost surely (Chin, 2021, Chin, 2019, Duy, 2014). The law admits natural generalizations to variable variance profiles and to ensembles with weak dependencies, provided sufficient control of cumulants or decay of row-wise variance deviations (Krajewski, 2017, Krajewski et al., 2016, Hochstättler et al., 2014).

The semicircle law is not limited to strictly independent entries; for example, ensembles with Curie–Weiss-type correlations exhibit semicircular limiting behavior below critical temperature, as long as the leading correlations decay sufficiently with the system size (Hochstättler et al., 2014).

2. Proof Schemes: Moment and Resolvent Methods

The two principal proof strategies are the method of moments and the resolvent (Stieltjes transform) approach (Duy, 2014, Benaych-Georges et al., 2016, Chin, 2019).

• Moment Method: One computes the expected trace of matrix powers, enumerating closed walks on the adjacency graph of indices. Only pairwise matchings (non-crossing pairings or Catalan structures) contribute at leading order due to centering and independence, leading to limiting moments corresponding to the Catalan numbers:

$$\frac{1}{n}\mathbb{E}\,\mathrm{Tr}\,W_n^{k} \to \begin{cases} \frac{1}{(k/2)+1}\binom{k}{k/2}, & k\text{ even} \ 0, & k\text{ odd} \end{cases}$$

which are precisely the moments of $\rho_{sc}$ (Duy, 2014, Kornyik et al., 2015, Chin, 2019).

• Resolvent/Stieltjes Transform: The normalized trace of the resolvent $G(z) = (W_n - zI)^{-1}$, $W_n$0, is shown to satisfy a self-consistent equation:

$W_n$1

whose unique solution in the upper half-plane is the Stieltjes transform of the semicircle density (Benaych-Georges et al., 2016, Krajewski, 2017, Chin, 2019). Concentration phenomena and high-probability estimates extend the convergence to local scales containing only a few eigenvalues (Erdos et al., 2010, Götze et al., 2016, Götze et al., 2015, Benaych-Georges et al., 2016).

3. Extensions: Local Laws, Universality, and Eigenvector Delocalization

The global law gives way to a local semicircle law when one considers the spectral density on intervals shrinking with $W_n$2:

• Local Law: The Stieltjes transform $W_n$3 remains within $W_n$4 of $W_n$5 for spectral parameters $W_n$6 with $W_n$7 down to $W_n$8. This enables precise control of the eigenvalue counting function on intervals $W_n$9 (Erdos et al., 2010, Götze et al., 2016, Götze et al., 2015, Benaych-Georges et al., 2016, Bao et al., 2011, Bauerschmidt et al., 2015).
• Eigenvector Delocalization: The local law implies that all eigenvectors are completely delocalized, i.e., the maximum entry of any normalized eigenvector is $\mathrm{Var}(W_{ij}) = 1/n$0 with high probability (Götze et al., 2016, Erdős et al., 2011, Bauerschmidt et al., 2015).
• Rigidity and Universality: Eigenvalues are shown to be rigid, i.e., $\mathrm{Var}(W_{ij}) = 1/n$1 is within $\mathrm{Var}(W_{ij}) = 1/n$2 (up to logarithmic factors) of its classical location, and the gap and correlation statistics in the bulk become universal—matching those of Gaussian ensembles—even when the fourth moment differs (Erdos et al., 2010, Götze et al., 2016, Benaych-Georges et al., 2016, Bao et al., 2011, Bauerschmidt et al., 2015, Erdős et al., 2011).

4. Necessary and Sufficient Conditions

The classical semicircle law requires normalization of the entry variances such that the row sums converge to unity. Precise necessary and sufficient conditions are given in terms of variance profile regularity (row sums deviating by $\mathrm{Var}(W_{ij}) = 1/n$3 from $\mathrm{Var}(W_{ij}) = 1/n$4), a Lindeberg condition controlling entry tails, and exclusion of vanishing mass in a sublinear fraction of rows (Chin, 2021, Chin, 2019). The semicircle limit persists under suitable domain-of-attraction and truncated-moment hypotheses, even when entries have infinite variance (Zhou, 2011).

Summary Table: Required Conditions for Semicircle Law

Condition Type	Typical Requirement	Source
Centering	$\mathrm{Var}(W_{ij}) = 1/n$5	(Chin, 2021, Duy, 2014)
Row sum of variances	$\mathrm{Var}(W_{ij}) = 1/n$6	(Chin, 2021)
Lindeberg/Uniform moment	$\mathrm{Var}(W_{ij}) = 1/n$7	(Chin, 2021, Chin, 2019)
No mass in $\mathrm{Var}(W_{ij}) = 1/n$8 rows	For $\mathrm{Var}(W_{ij}) = 1/n$9: $i \neq j$0	(Chin, 2021)

5. Beyond Classic Ensembles: Regular Graphs, Networks, and Tensors

The Wigner semicircle law extends to other matrix models with more complex structure:

• Random Regular Graphs: The rescaled adjacency matrix of random $i \neq j$1-regular graphs with $i \neq j$2 displays the semicircle law in both global and local forms, despite the strong dependency among entries (Bauerschmidt et al., 2015). Delocalization and spectral gap results are achieved via a generalization of the Green function approach.
• Erdős–Rényi Graphs and Weighted Networks: The normalized Laplacian of weighted networks or random graphs converges to a semicircular law under minimal degree and moment constraints, allowing spectral predictions from summary statistics (e.g., average degree and second moment of weights) (Sakumoto et al., 2020, Erdős et al., 2011).
• Generalized Tensors: For random symmetric tensors (order $i \neq j$3), the limiting spectral law generalizes the semicircle law through the Fuss–Catalan distribution, with the classical law recovered for $i \neq j$4 (Gurau, 2020).

6. Semicircle Law in Random Polynomials and Hermite Zeros

Remarkably, the semicircle law governs the asymptotic root statistics of certain random polynomials. For random real-rooted polynomials whose roots are i.i.d., repeated differentiation followed by scaling yields zeros distributed according to the semicircle law; the connection is explicit via convergence to the Hermite polynomial zeros, which themselves normalize to the semicircle (Hoskins et al., 2020, Kornyik et al., 2015).

7. Robustness and Technical Extensions

The law endures in the presence of weak dependencies, block structures, or decaying correlations, provided cumulant bounds analogous to Gaussian ensembles are enforced (Krajewski, 2017, Krajewski et al., 2016, Hochstättler et al., 2014). For real symmetric or Hermitian matrices with dependent but suitably controlled entries, the renormalization group or effective action methods validate semicircular convergence by suppressing non-Gaussian contributions in the large $i \neq j$5 limit (Krajewski, 2017, Krajewski et al., 2016).

• "Necessary and Sufficient Conditions for Convergence to the Semicircle Distribution" (Chin, 2021)
• "An Exposition on Wigner's Semicircular Law" (Chin, 2019)
• "Central limit theorem for moments of spectral measures of Wigner matrices" (Duy, 2014)
• "Lectures on the local semicircle law for Wigner matrices" (Benaych-Georges et al., 2016)
• "On the Local Semicircular Law for Wigner Ensembles" (Götze et al., 2016)
• "A renormalisation group approach to the universality of Wigner's semicircle law for random matrices with dependent entries" (Krajewski, 2017)
• "Wigner law for matrices with dependent entries - a perturbative approach" (Krajewski et al., 2016)
• "Semicircle law for a matrix ensemble with dependent entries" (Hochstättler et al., 2014)
• "Rigidity of Eigenvalues of Generalized Wigner Matrices" (Erdos et al., 2010)
• "Local Semicircle law and Gaussian fluctuation for Hermite $i \neq j$6 ensemble" (Bao et al., 2011)
• "Wigner matrices, the moments of roots of Hermite polynomials and the semicircle law" (Kornyik et al., 2015)
• "A Semicircle Law for Derivatives of Random Polynomials" (Hoskins et al., 2020)
• "On the generalization of the Wigner semicircle law to real symmetric tensors" (Gurau, 2020)
• "Local semicircle law for random regular graphs" (Bauerschmidt et al., 2015)
• "The Wigner's Semicircle Law of Weighted Random Networks" (Sakumoto et al., 2020)
• "Spectral statistics of Erdős-Rényi graphs I: Local semicircle law" (Erdős et al., 2011)
• "New estimators of spectral distributions of Wigner matrices" (Zhou, 2011)
• "Local semicircle law under moment conditions. Part I: The Stieltjes transform" (Götze et al., 2015)